CS 201 | Hossein Mobahi, Google DeepMind

Avoiding Spurious Sharpness Minimization

Abstract:
Curvature regularization techniques like Sharpness Aware Minimization (SAM) have shown great promise in improving generalization on vision tasks. However, we find that SAM performs poorly in domains like natural language processing (NLP), often degrading performance — even with twice the compute budget. We investigate the discrepancy across domains and find that in the NLP setting, SAM is dominated by regularization of the logit statistics — instead of improving the geometry of the function itself. We use this observation to develop an alternative algorithm we call Functional SAM, which regularizes curvature only through modification of the statistics of the overall function implemented by the neural network, and avoids spurious minimization through logit manipulation. Furthermore, we argue that preconditioning the SAM perturbation also prevents spurious minimization, and when combined with Functional SAM, it gives further improvements. Our proposed algorithms show improved performance over AdamW and SAM baselines when trained for an equal number of steps, in both fixed-length and Chinchilla-style training settings, at various model scales (including billion-parameter scale). On the whole, our work highlights the importance of more precise characterizations of sharpness in broadening the applicability of curvature regularization to large language models (LLMs).

Bio:
Hossein Mobahi is a senior research scientist at Google DeepMind. His research interests center on the interplay between optimization and generalization in deep neural networks. He is one of the creators of the Sharpness-Aware Minimization (SAM) algorithm, which has been widely recognized by the community as a major tool for improving generalization performance. Before joining Google in 2016, he was a postdoctoral researcher at MIT’s CSAIL. He received his PhD in Computer Science from the University of Illinois at Urbana-Champaign (UIUC).

Date/Time:
Date(s) - Feb 18, 2025
4:00 pm - 5:45 pm

Location:
3400 Boelter Hall
420 Westwood Plaza Los Angeles California 90095